Dear my Brazillion new followers, I really don't want to disappoint, but I think you should know that it looks like RIO 397b might've just been an effort to embezzle funds.
Timeline:
1. Model training received funding of R$500K (~$100K USD).
2. Initial model documentation claimed that it was a developed on top of Qwen 3.5 397b with further training and technological improvements.
3. The model was a simple merge of 40% Nex N2 Pro and 60% Qwen 3.5 without further training.
4. Model card readme was updated to admit that it was an Nex N2 Pro merge, while maintaining that additional training still took place, and they simply uploaded the wrong model. Then the previously uploaded model was removed from HF.
5. The final model got lost, "We tried to recover the final model, but it was not possible. It will be released only after the new training and all external validations are completed." So they'll have to redo it from scratch.
Taken as a whole, it reads like "We pocketed funding, delivered a fake result, were caught, and now promise to do the actual work."
Citations in thread 👇
Don't hate the messenger, but if I'm getting something wrong, do let me know below.
Dear my Brazillion new followers, I really don't want to disappoint, but I think you should know that it looks like RIO 397b might've just been an effort to embezzle funds.
Timeline:
1. Model training received funding of R$500K (~$100K USD).
2. Initial model documentation claimed that it was a developed on top of Qwen 3.5 397b with further training and technological improvements.
3. The model was a simple merge of 40% Nex N2 Pro and 60% Qwen 3.5 without further training.
4. Model card readme was updated to admit that it was an Nex N2 Pro merge, while maintaining that additional training still took place, and they simply uploaded the wrong model. Then the previously uploaded model was removed from HF.
5. The final model got lost, "We tried to recover the final model, but it was not possible. It will be released only after the new training and all external validations are completed." So they'll have to redo it from scratch.
Taken as a whole, it reads like "We pocketed funding, delivered a fake result, were caught, and now promise to do the actual work."
Citations in thread 👇
Don't hate the messenger, but if I'm getting something wrong, do let me know below.
The Rio 3.5 model broke the internet this week. The plot twist? It’s essentially our open-source model, Nex N2 Pro, wearing a different hat.
🤯 We analyzed the weights, and the recipe is exact: Rio 3.5 ≈ 0.6 * Nex N2 Pro + 0.4 * Qwen 3.5
It even literally introduces itself as "Nex N2 Pro" if you ask it without initial system prompt!
😂 We are flattered that the City of Rio used our work to achieve SOTA performance. Thanks for the ultimate benchmark validation.
🤝 But in the open-source world, attribution matters.
👇 Full mathematical proof & verify script in the first reply!
É absolutamente INACREDITÁVEL ninguém da comissão técnica perceber que Casemiro e Raphinha não possuem a MENOR condição de serem titulares na seleção mais vencedora do mundo. É surreal isso. Surreal.
- O Bolsonaro é o culpado pela farra do INSS
- Bora investigar então?
- Não
- O Bolsonaro é culpado pelo escândalo do Master
- Bora assinar a CPMI então?
- Não
- O Bolsonaro tem ligação com o CV e PCC
- Bora classificar como organizações terroristas então?
- Não
Grok foundation model V9-Medium (1.5T) has finished training. Evals look good. A lot of Cursor data was added in supplementary training and there is more to come.
Fine-tuning is underway and reinforcement learning begins in a few days. 2 to 3 weeks to public release.
This will be a major improvement over the 0.5T v8-small that currently serves all Grok production traffic, especially for difficult coding tasks.